| Definition |
A type of learning where the model is trained on a labeled dataset, which means that the training data includes both input data and the corresponding correct outputs. |
Learning from an unlabeled dataset, the model tries to find the underlying patterns and structures in the data. |
A type of learning where the model learns to interact with an environment to achieve a goal or maximize some notion of cumulative reward. |
| Training Data |
Labeled data (features and labels) |
Unlabeled data (features only) |
Interaction with the environment, rewards based on actions. |
| Goal |
To make accurate predictions or classifications based on the input data. |
To find hidden patterns or groupings in the data. |
To find a strategy to obtain the maximum cumulative reward over time. |
| Algorithms |
Decision Trees, Support Vector Machines, Neural Networks, etc. |
Clustering (e.g., K-means), Association (e.g., Apriori), Principal Component Analysis, etc. |
Q-learning, Deep Q Network (DQN), Policy Gradients, etc. |
| Real-world Applications |
Image recognition, Spam detection, Credit risk analysis, etc. |
Market segmentation, Anomaly detection, Recommender systems, etc. |
Autonomous vehicles, Game playing (like AlphaGo), Robotics, etc. |
| Evaluation Metrics |
Accuracy, Precision, Recall, F1-score, etc. |
Silhouette score, Davies-Bouldin index, etc. |
Reward function, which may vary greatly depending on the specific task. |